* previous encounters. It's functional, and as neat as it can be in the
* circumstances, but be wary, for these things are subtle and break easily.
* The Guest provides a virtual to physical mapping, but we can neither trust
- * it nor use it: we verify and convert it here to point the hardware to the
- * actual Guest pages when running the Guest. :*/
+ * it nor use it: we verify and convert it here then point the CPU to the
+ * converted Guest pages when running the Guest. :*/
/* Copyright (C) Rusty Russell IBM Corporation 2006.
* GPL v2 and any later version */
#include <linux/percpu.h>
#include <asm/tlbflush.h>
#include <asm/uaccess.h>
+#include <asm/bootparam.h>
#include "lg.h"
/*M:008 We hold reference to pages, which prevents them from being swapped.
BUG_ON(!(pgd_flags(gpgd) & _PAGE_PRESENT));
return gpage + ((vaddr>>PAGE_SHIFT) % PTRS_PER_PTE) * sizeof(pte_t);
}
+/*:*/
+
+/*M:014 get_pfn is slow: we could probably try to grab batches of pages here as
+ * an optimization (ie. pre-faulting). :*/
/*H:350 This routine takes a page number given by the Guest and converts it to
* an actual, physical page number. It can fail for several reasons: the
* and the page is read-only, or the write flag was set and the page was
* shared so had to be copied, but we ran out of memory.
*
- * This holds a reference to the page, so release_pte() is careful to
- * put that back. */
+ * This holds a reference to the page, so release_pte() is careful to put that
+ * back. */
static unsigned long get_pfn(unsigned long virtpfn, int write)
{
struct page *page;
- /* This value indicates failure. */
- unsigned long ret = -1UL;
- /* get_user_pages() is a complex interface: it gets the "struct
- * vm_area_struct" and "struct page" assocated with a range of pages.
- * It also needs the task's mmap_sem held, and is not very quick.
- * It returns the number of pages it got. */
- down_read(¤t->mm->mmap_sem);
- if (get_user_pages(current, current->mm, virtpfn << PAGE_SHIFT,
- 1, write, 1, &page, NULL) == 1)
- ret = page_to_pfn(page);
- up_read(¤t->mm->mmap_sem);
- return ret;
+ /* gup me one page at this address please! */
+ if (get_user_pages_fast(virtpfn << PAGE_SHIFT, 1, write, &page) == 1)
+ return page_to_pfn(page);
+
+ /* This value indicates failure. */
+ return -1UL;
}
/*H:340 Converting a Guest page table entry to a shadow (ie. real) page table
/*H:460 And to complete the chain, release_pte() looks like this: */
static void release_pte(pte_t pte)
{
- /* Remember that get_user_pages() took a reference to the page, in
+ /* Remember that get_user_pages_fast() took a reference to the page, in
* get_pfn()? We have to put it back now. */
if (pte_flags(pte) & _PAGE_PRESENT)
put_page(pfn_to_page(pte_pfn(pte)));
* all processes. So when the page table above that address changes, we update
* all the page tables, not just the current one. This is rare.
*
- * The benefit is that when we have to track a new page table, we can copy keep
- * all the kernel mappings. This speeds up context switch immensely. */
+ * The benefit is that when we have to track a new page table, we can keep all
+ * the kernel mappings. This speeds up context switch immensely. */
void guest_set_pte(struct lg_cpu *cpu,
unsigned long gpgdir, unsigned long vaddr, pte_t gpte)
{
- /* Kernel mappings must be changed on all top levels. Slow, but
- * doesn't happen often. */
+ /* Kernel mappings must be changed on all top levels. Slow, but doesn't
+ * happen often. */
if (vaddr >= cpu->lg->kernel_address) {
unsigned int i;
for (i = 0; i < ARRAY_SIZE(cpu->lg->pgdirs); i++)
release_pgd(lg, lg->pgdirs[pgdir].pgdir + idx);
}
+/* Once we know how much memory we have we can construct simple identity
+ * (which set virtual == physical) and linear mappings
+ * which will get the Guest far enough into the boot to create its own.
+ *
+ * We lay them out of the way, just below the initrd (which is why we need to
+ * know its size here). */
+static unsigned long setup_pagetables(struct lguest *lg,
+ unsigned long mem,
+ unsigned long initrd_size)
+{
+ pgd_t __user *pgdir;
+ pte_t __user *linear;
+ unsigned int mapped_pages, i, linear_pages, phys_linear;
+ unsigned long mem_base = (unsigned long)lg->mem_base;
+
+ /* We have mapped_pages frames to map, so we need
+ * linear_pages page tables to map them. */
+ mapped_pages = mem / PAGE_SIZE;
+ linear_pages = (mapped_pages + PTRS_PER_PTE - 1) / PTRS_PER_PTE;
+
+ /* We put the toplevel page directory page at the top of memory. */
+ pgdir = (pgd_t *)(mem + mem_base - initrd_size - PAGE_SIZE);
+
+ /* Now we use the next linear_pages pages as pte pages */
+ linear = (void *)pgdir - linear_pages * PAGE_SIZE;
+
+ /* Linear mapping is easy: put every page's address into the
+ * mapping in order. */
+ for (i = 0; i < mapped_pages; i++) {
+ pte_t pte;
+ pte = pfn_pte(i, __pgprot(_PAGE_PRESENT|_PAGE_RW|_PAGE_USER));
+ if (copy_to_user(&linear[i], &pte, sizeof(pte)) != 0)
+ return -EFAULT;
+ }
+
+ /* The top level points to the linear page table pages above.
+ * We setup the identity and linear mappings here. */
+ phys_linear = (unsigned long)linear - mem_base;
+ for (i = 0; i < mapped_pages; i += PTRS_PER_PTE) {
+ pgd_t pgd;
+ pgd = __pgd((phys_linear + i * sizeof(pte_t)) |
+ (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER));
+
+ if (copy_to_user(&pgdir[i / PTRS_PER_PTE], &pgd, sizeof(pgd))
+ || copy_to_user(&pgdir[pgd_index(PAGE_OFFSET)
+ + i / PTRS_PER_PTE],
+ &pgd, sizeof(pgd)))
+ return -EFAULT;
+ }
+
+ /* We return the top level (guest-physical) address: remember where
+ * this is. */
+ return (unsigned long)pgdir - mem_base;
+}
+
/*H:500 (vii) Setting up the page tables initially.
*
* When a Guest is first created, the Launcher tells us where the toplevel of
* its first page table is. We set some things up here: */
-int init_guest_pagetable(struct lguest *lg, unsigned long pgtable)
+int init_guest_pagetable(struct lguest *lg)
{
+ u64 mem;
+ u32 initrd_size;
+ struct boot_params __user *boot = (struct boot_params *)lg->mem_base;
+
+ /* Get the Guest memory size and the ramdisk size from the boot header
+ * located at lg->mem_base (Guest address 0). */
+ if (copy_from_user(&mem, &boot->e820_map[0].size, sizeof(mem))
+ || get_user(initrd_size, &boot->hdr.ramdisk_size))
+ return -EFAULT;
+
/* We start on the first shadow page table, and give it a blank PGD
* page. */
- lg->pgdirs[0].gpgdir = pgtable;
+ lg->pgdirs[0].gpgdir = setup_pagetables(lg, mem, initrd_size);
+ if (IS_ERR_VALUE(lg->pgdirs[0].gpgdir))
+ return lg->pgdirs[0].gpgdir;
lg->pgdirs[0].pgdir = (pgd_t *)get_zeroed_page(GFP_KERNEL);
if (!lg->pgdirs[0].pgdir)
return -ENOMEM;
/* We've made it through the page table code. Perhaps our tired brains are
* still processing the details, or perhaps we're simply glad it's over.
*
- * If nothing else, note that all this complexity in juggling shadow page
- * tables in sync with the Guest's page tables is for one reason: for most
- * Guests this page table dance determines how bad performance will be. This
- * is why Xen uses exotic direct Guest pagetable manipulation, and why both
- * Intel and AMD have implemented shadow page table support directly into
- * hardware.
+ * If nothing else, note that all this complexity in juggling shadow page tables
+ * in sync with the Guest's page tables is for one reason: for most Guests this
+ * page table dance determines how bad performance will be. This is why Xen
+ * uses exotic direct Guest pagetable manipulation, and why both Intel and AMD
+ * have implemented shadow page table support directly into hardware.
*
* There is just one file remaining in the Host. */
projects / mv-sheeva.git / blobdiff